The Toll Booth

Last week I uploaded a track to my distributor and the form asked me a question: Does this song include AI-generated music, vocals, or lyrics? Directly beneath it, in smaller text, a reassurance: Using AI for mastering or mixing doesn't count.

Read that twice. AI that shapes every frequency of the final record you hear — doesn't count. AI that renders a performance — counts. Somebody drew that line. Nobody voted on it. And the placement of that line is the most honest document the music industry has produced in years, because it maps, with surgical precision, onto exactly one thing: whose invoice gets threatened.

I want to take the arguments for that line seriously — more seriously than the people making them usually do. I'm going to give each one its full weight, and then I'm going to walk it forward until it collapses. Every one of them collapses. What's left standing at the end is one honest fact and one toll booth.

First, the pattern, because none of this is new.

When the drum machine arrived, it wasn't real drumming. When the synthesizer arrived, the UK Musicians' Union campaigned against it under the banner of keeping music live — the machine was going to kill musicianship. When sampling arrived, it was theft, right up until the companies condemning it owned the publishing on the samples, at which point it became a licensing revenue stream. When Auto-Tune arrived, it was cheating; today it is on functionally every commercial vocal record, including the ones made by the people drawing the current line. In 1942 the American Federation of Musicians went further than anyone: it banned its members from making records at all, because recorded music was destroying the livelihoods of live players — which, to be fair, it was. The cinemas and hotels and radio stations that once employed orchestras full of working musicians replaced them with playback. The industry that did that replacing is the same industry now explaining that certain technology crosses a moral line.

The line has never, in a century of this pattern, been drawn at "what is art." It is drawn, every single time, at whose workflow is already inside the fence. AI mastering doesn't count because the industry adopted algorithmic mastering years ago; it lives inside the fence. Generative production threatens the labor structure the fence was built to protect. That's the whole distinction. It has a costume — aesthetics, authenticity, the soul of music — but underneath the costume it is an employment map.

If you doubt that, look at the industry's own brand-new taxonomy. The major industry groups have now rolled out a global labeling system: a work is merely "AI-assisted" — admissible, respectable — if it has human lead vocals and primary instruments. Sit with that definition. It is not a statement about creativity, intention, authorship, or artistic merit. It is a staffing requirement. A composition conceived, structured, and written entirely by a machine qualifies as human music if humans were paid to perform it. A composition conceived, structured, and written entirely by a human is machine music if a machine performed it. The definition doesn't protect art. It protects payroll.

Now the counterarguments — at full strength.

The provenance argument. This is the sturdy one, and it deserves respect: the generative models were trained on enormous corpora of copyrighted music without permission. That's a genuine ethical objection, different in kind from incumbent line-drawing — a mastering algorithm wasn't built on a contested dataset of other people's catalogs. I grant all of it.

Now watch it fall twice. First, empirically: the objection is dissolving on a schedule. The lawsuits produced settlements; the settlements produced licensing deals; the licensed models are replacing the unlicensed ones. An ethical objection with an expiration date and a price was never a principle — it was a negotiation. The industry didn't fight to stop the machines; it fought to get paid by them, got paid, and stopped fighting. Second, by symmetry: every human musician who has ever lived trained on a corpus of copyrighted work they didn't license. Influence is ingestion. Nobody who ever wrote a song paid royalties to everyone they absorbed, and the industry never dreamed of asking them to. Examine the provenance objection closely and it isn't "training on unlicensed work is wrong" — humans have always done that. It's "training at machine scale is wrong." Which is not an ethics claim. It's an objection to throughput, invoiced as ethics. The toll booth again.

The economics argument. Session players, vocalists, working musicians — real people with real mortgages are watching a machine do their job for pennies. This is true, and it deserves compassion without any irony attached. It is also, as an argument for gatekeeping, null — and the industry making it has a century of its own precedent proving it null. Every technology displaces the labor it automates. The synthesizer displaced string sections. The drum machine displaced drummers. Recorded music itself displaced more working musicians than every technology since combined, and the industry now claiming moral authority is the one that did it. Compassion for the displaced has always been owed. Veto power for the displaced has never once been granted — least of all by the labels, platforms, and rights organizations that profited from every previous displacement. They are not defending musicians. They are defending this year's payroll structure, having shredded the last three without a backward glance.

The authenticity argument. Some listeners genuinely want a human being behind the sound — want to believe someone sweated, struggled, lived the thing they're hearing. This one survives examination. But look carefully at what survives: a market preference, not a moral principle. Some real fraction of the audience prefers the human-sweat story, and that demand would exist without any institution enforcing it. Fine. Now notice that the industry claiming to defend that authenticity has spent a hundred years manufacturing it: uncredited ghostwriters behind beloved singer-songwriters, session musicians behind band photos, careers built on personas engineered in conference rooms. Milli Vanilli remains the perfect case, because notice precisely what their crime was — not the production method, which was industry-standard, but the disclosure failure. The audience's authenticity preference is a preference for a story, and stories are marketing. A marketing preference can shape what sells. It cannot justify a gate on what is admissible. And there is a clean answer to it that requires no argument at all: listeners who choose the music before they know or care how it was made. My catalog took sixty-two thousand plays in its first six weeks of existence, organically, with zero promotional spend. Nobody who pressed play was sold a sweat story. They voted on the artifact.

Which brings me to the claim underneath all three counterarguments — the quiet assumption that the machine is the music, that anyone with the subscription makes the song. Test it. Hand a stranger the same tools I use and watch what happens: they produce what the platforms are drowning in — the flood one major streaming service reports as nearly half its daily uploads, and calls slop. The tool is uniform. The output distribution is not. The distance between the slop pile and a catalog that converts — thirteen tracks signed to a label that keeps asking for more — is everything that doesn't ship with the subscription: the production doctrine built from hundreds of failed generations, the arrangement discipline, the vocal architectures, the verification gates, the taste to know which of four renders lives and which three die. That corpus is technique. It was acquired the way technique has always been acquired — iteration, failure, correction, codification. The instrument changed. The ten thousand hours moved; they did not disappear.

And this, too, has always been true of composition. Georg Cretu built Enigma — seventy million records — by looping borrowed voices over a drum pattern in a home studio; the musicianship was curation and architecture. Irving Berlin, arguably the most successful songwriter in American history, could not read or write notation; he hummed to a transcriber. Hans Zimmer's scores reach the orchestra as machine mockups passed through orchestrators. The composer has always been, functionally, the prompt engineer of the ensemble — the person who specifies, in whatever language reaches the performers, exactly what must be played. Composition and performance were never the same skill. The industry has always run on their separation. What changed is that the performance layer got cheap, and the people who sell performance are calling their price collapse a moral crisis.

Here is the thought experiment that ends the debate, and I invite anyone defending the current line to sit with it. My production pipeline will, before long, emit conventional sheet music. Suppose I walk into any studio in the country with that notation and hire human players — union rates, full credits — to perform it. A recording now exists with human lead vocals and human primary instruments. Under the industry's own taxonomy, published this year, that record is AI-assisted: fully admissible. I could disclose the entire pipeline at the door and still walk through it — because the fence was never built around AI. It was built around whether my money touches their labor. The identical composition is machine slop or human art depending on one variable: who got paid to render it.

That is not an aesthetic principle. That is a toll booth.

So here is where I land, stated plainly. I check Yes on the form — every release, every time — not because the line deserves respect but because my position doesn't need the lie; disclosure is the high ground and I intend to keep it. I take the two honest residues seriously: provenance mattered and is being resolved with licenses, and some listeners will always prefer the sweat story, which is their right and my marketing problem, not my moral one. And I am not asking to join the club. The gates are guarding a shrinking territory: music can now be conceived, written, and produced faster, cheaper, and — in the hands of anyone willing to do the real work of learning the new instrument — with more range than the fenced-in model can match. Every wall they build raises the value of everything built on this side of it.

They should be worried. Not about the machines — about the fact that the machines revealed what the fences were always for.

Wayne M. Kirkman III is a composer, producer, and the founder of Ex Pulvis Records, an independent label. All catalog figures cited are his own, and yes, he checked the box.